Scripts not exiting

[mhart]

The following script prints done but doesn't exit. On aws-sdk 2.0.3 and node v0.10.29:

var AWS = require('aws-sdk')
var ec2 = new AWS.EC2

ec2.describeInstances(function() {
  console.log('done')
})

Perhaps there's something still being held by the event loop?

This may also be related to #307?

[lsegal]

This is not related to #307. This is related to the use of the new keepalive agent being used by the SDK to improve latency when sending many requests to a service. It looks like this agent is keeping the connection alive indefinitely. I'm investigating if there is an issue with the library of if it is just being misused. You can workaround this issue by reverting back to the global agent for now:

AWS.config.httpOptions = {agent: require('https').globalAgent};

[mhart]

Using agentkeepalive as described on its README seems to work fine (ie, the example exits ok) - but I guess there may be issues with more advanced usage...?

[mhart]

Hmmm, I didn't try https though

[lsegal]

Using the httpOptions.agent parameter should be equivalent to using it in the context of a low-level https.request call. All we do is pass that agent object through to the options if one was provided (if one was not, we provide a default): https://github.com/aws/aws-sdk-js/blob/master/lib/http/node.js#L35-L39

In other words, if you passed an agentkeepalive Agent.HttpsAgent() in, it should be equivalent to the example. Something else must be going on that causes the hang, but it is somehow related to this library.

[mhart]

Just tried, their https example works fine too... Not sure what's going on!

[lsegal]

It has to do with timeout behavior on the socket. Adding req.setTimeout(0); prior to req.end() in the agentkeepalive examples reproduces the behavior. The issue is that agentkeepalive seems to rely on timeout behavior and the SDK is resetting this timeout behavior. I'm going to keep investigating to see if there's any way I can make this work transparently, but I'm thinking that we may have to pull agentkeepalive out until we can figure out a better way to get this behavior working.

[lsegal]

I've reverted agentkeepalive for the time being until we can figure out a less intrusive way to add keepalive behavior. This should be out in the next release.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs and link to relevant comments in this thread.